Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH] x86/boot: drop memset from copy.S
To:     <hpa@zytor.com>, <x86@kernel.org>, <linux-kernel@vger.kernel.org>
CC:     <tglx@linutronix.de>, <mingo@redhat.com>, <bp@alien8.de>
References: <20190107074056.12532-1-caoj.fnst@cn.fujitsu.com>
 <000CE7C8-33F3-404A-9E7D-F2CCB0BA110E@zytor.com>
 <28158fd7-8569-e0c1-75d0-c423330e256d@cn.fujitsu.com>
 <1DA6B342-5F32-4C73-9352-4DB9F176D53C@zytor.com>
From:   Cao jin <caoj.fnst@cn.fujitsu.com>
Message-ID: <181a8787-fb13-2c88-2d85-00c1bb377cdf@cn.fujitsu.com>
Date:   Thu, 10 Jan 2019 16:28:38 +0800
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.2.1
MIME-Version: 1.0
In-Reply-To: <1DA6B342-5F32-4C73-9352-4DB9F176D53C@zytor.com>
Content-Type: text/plain; charset="utf-8"
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hello HPA,

On 1/8/19 4:38 PM, hpa@zytor.com wrote:
> On January 7, 2019 12:52:57 AM PST, Cao jin <caoj.fnst@cn.fujitsu.com> wrote:
>> Hi,
>>
>> On 1/7/19 3:59 PM, hpa@zytor.com wrote:
>>> On January 6, 2019 11:40:56 PM PST, Cao jin
>> <caoj.fnst@cn.fujitsu.com> wrote:
>>>> According to objdump output of setup, function memset is not used in
>>>> setup code. Currently, all usage of memset in setup come from macro
>>>> definition of string.h.
>>>>
>>>> Signed-off-by: Cao jin <caoj.fnst@cn.fujitsu.com>
>>>> ---
>>>> Compiled and booted under x86_64; compiled under i386.
>>>>
>>>> Questions: now there is 2 definition of memcpy, one lies in copy.S,
>>>> another lies in string.h which is mapped to gcc builtin function. Do
>> we
>>>> still need 2 definition? Could we move the content of copy.S to
>>>> boot/string.c?
>>>>
>>>> At first glance, the usage of string.{c.h} of setup is kind of
>>>> confusing,
>>>> they are also used in compressed/ and purgatory/
>>>>
>>>> arch/x86/boot/copy.S | 15 ---------------
>>>> 1 file changed, 15 deletions(-)
>>>>
>>>> diff --git a/arch/x86/boot/copy.S b/arch/x86/boot/copy.S
>>>> index 15d9f74b0008..5157d08b0ff2 100644
>>>> --- a/arch/x86/boot/copy.S
>>>> +++ b/arch/x86/boot/copy.S
>>>> @@ -33,21 +33,6 @@ GLOBAL(memcpy)
>>>> 	retl
>>>> ENDPROC(memcpy)
>>>>
>>>> -GLOBAL(memset)
>>>> -	pushw	%di
>>>> -	movw	%ax, %di
>>>> -	movzbl	%dl, %eax
>>>> -	imull	$0x01010101,%eax
>>>> -	pushw	%cx
>>>> -	shrw	$2, %cx
>>>> -	rep; stosl
>>>> -	popw	%cx
>>>> -	andw	$3, %cx
>>>> -	rep; stosb
>>>> -	popw	%di
>>>> -	retl
>>>> -ENDPROC(memset)
>>>> -
>>>> GLOBAL(copy_from_fs)
>>>> 	pushw	%ds
>>>> 	pushw	%fs
>>>
>>> This is dependent on both gcc version and flags.
>>>
>>
>> Thanks for your info, but I still don't quite get your point. All files
>> who has memset reference in setup will also #include "string.h", so how
>> gcc version and flags will associate memset reference to the assembly
>> function by force?  Hope to get a little more details when you are
>> convenient:)
> 
> GCC will sometimes emit calls to certain functions including memcpy().
> 

Thanks very much. After spending some time on GCC document, I think you
are talking about a condition that, for example, __builtin_memcpy() is
not optimized as inline code, but a function call to memcpy() in copy.S.

But I failed to find out the details how would gcc version & flags make
it this way, even I checked out the .cmd file of these .c. Or is this
born to be obscure for programmers?

-- 
Sincerely,
Cao jin